Dataset updated 2026-10-24

Flint 40B-A2.5B

Flint 40B-A2.5B loads in 22.9 GB at NVFP4 and decodes at the speed of its 2.5B active path, comfortably from 36 GB of memory, Macs included, starting at the MacBook Pro M5 Pro 48GB. Below is the memory math we ran for it, the smallest card that fits, and the fastest machine we track.

Parameters
39.8B(2.5B active)
Format
NVFP4
Min memory
36 GB
Best for
Agentic coding, Long context
Memory math

How much memory does Flint 40B-A2.5B need?

Flint 40B-A2.5B is a 39.8B mixture-of-experts model (2.5B active per token) built for agentic coding, long context. The weights alone take 22.9 GB at NVFP4; add context and the OS, and it belongs in the 36 GB memory class. Every number below comes from the ModelFit engine: speeds are estimates, labeled est., because we do not benchmark hardware ourselves.

Weights (NVFP4)
22.9 GB
+ KV cache (16k)
~0.4 GB
Total at 16k
~23.3 GB
Comfortable from
36 GB

KV = fp16 estimate (q8_0 cache roughly halves it). "Comfortable" = weights + KV within the engine's tiered budget (~70-85% of memory).

Hardware snapshot

Which machines run Flint 40B-A2.5B?

Smallest GPU
NVIDIA GeForce RTX 5090est. 363 tok/s · 32 GB VRAMtypical error 15%
Fastest
NVIDIA GeForce RTX 5090est. 363 tok/stypical error 15%
Runs on a Mac?
MacBook Pro M5 Pro 48GBest. 107 tok/stypical error 10%

Verdicts come from the same engine that powers our fit checker: memory capacity first, bandwidth for speed. How we compute these numbers · What typical error means · Hugging Face model card

Frequently asked questions

How much memory does Flint 40B-A2.5B need?

22.9 GB for the NVFP4 weights, plus ~0.4 GB of KV-cache at 16k context, about 23.3 GB total. Comfortable from 36 GB of VRAM or unified memory.

Does Flint 40B-A2.5B run on a Mac?

Yes, from the MacBook Pro M5 Pro 48GB (est. 107 tok/s, typical error 10%). Unified memory means the RAM budget is the only limit.

What is the smallest GPU for Flint 40B-A2.5B?

The NVIDIA GeForce RTX 5090 (32 GB VRAM) is the smallest tracked card that runs Flint 40B-A2.5B comfortably, est. 363 tok/s, typical error 15%.

Cite this page

ModelFit: Flint 40B-A2.5B, specs, memory math and hardware verdicts.
https://modelfit.io/models/flint-40b/ (dataset updated 2026-10-24, CC BY 4.0).